Getting error for LSTM training

[rish27]

HashtagGeneration/onmt/Models.py

Line 321 in 9d281db

hidden = self.combine_hidden(torch.cat([src_hidden, ans_hidden], dim=-1)) # [2, batch_size, hidden]

When configured on train using LSTM's getting below error:
Traceback (most recent call last):
File "../train.py", line 516, in
main()
File "../train.py", line 508, in main
train_model(model, fields, optim, data_type, model_opt)
File "../train.py", line 262, in train_model
train_stats = trainer.train(train_iter, epoch, opt.max_src_len, opt.max_conv_len, report_func)
File "/content/drive/My Drive/Colab Notebooks/IREProject/HashtagGeneration/onmt/Trainer.py", line 183, in train
max_src_len, max_conv_len)
File "/content/drive/My Drive/Colab Notebooks/IREProject/HashtagGeneration/onmt/Trainer.py", line 333, in _gradient_accumulation
max_src_len, max_conv_len, dec_state)
File "/usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py", line 541, in call
result = self.forward(*input, **kwargs)
File "/content/drive/My Drive/Colab Notebooks/IREProject/HashtagGeneration/onmt/Models.py", line 760, in forward
enc_final, memory_bank = self.encoder(src, conversation, lengths)
File "/usr/local/lib/python3.6/dist-packages/torch/nn/modules/module.py", line 541, in call
result = self.forward(*input, **kwargs)
File "/content/drive/My Drive/Colab Notebooks/IREProject/HashtagGeneration/onmt/Models.py", line 322, in forward
hidden = self.combine_hidden(torch.cat([src_hidden, ans_hidden], dim=-1)) # [2, batch_size, hidden]
TypeError: expected Tensor as element 0 in argument 0, but got tuple
TypeError: expected Tensor as element 0 in argument 0, but got tuple

[yuewang-cuhk]

Hi, I have not tested the LSTM cell yet as I adopt GRU cell for this project. So you should be careful when using LSTM cell (GRU is highly recommended). From this error prompts, we can see that the errors come from the different outputs of LSTM cell compared to GRU. For LSTM cell, it outputs a tuple of (hidden state h_n, cell state c_n). You need to convert this tuple into a Tensor and be careful about the dimension size when initialing decoder state (see self._bridge)